Handwritten Number Recognition Of NAO Robot Based On Neural Network

The NAO robot is a standardized platform for RoboCup.It has very good applicability and scalability.It can perform very rich tasks and has excellent performance in various types of work,but there are also many problems in it.In image recognition,NAO can only identify a specific logo or perform simple two-dimensional image entry and recognition,and the recognition rate is also unsatisfactory.In order to solve the limitations of the NAO robot identification icon content,as well as the recognition rate,the image recognition module is redesigned based on the neural network based on the NAO robot platform.This article first introduces the preparatory work required for some self-development of NAO robots,because the official API of NAO is not completely open source,we need to reconfigure the environment with our own needs,build some database and experimental platform,for which I Do a lot of preparation work.Secondly,after the image is acquired,the image is preprocessed to make it easier to be recognized,and at the same time,the amount of data fed into the network is reduced as much as possible.After that is the construction of the neural network.I first built a very classic BP neural network with relatively simple structure.BP neural network can accomplish many relatively simple tasks and has basically satisfactory experimental results.Therefore,my first experiment chose to use BP neural network.Identify.Secondly,it is a convolutional neural network.It is the most mainstream image recognition network at present.It has very good recognition ability and development potential.Many more complex networks in the field of image recognition are evolved from simple volumes and networks.In this paper,I mainly solved some problems in the construction of NAO robots in Linux environment,as well as the realization of various networks,and the problem of adapting network parameters to the actual situation of NAO robots.